That seems like a horrible core idea. How is that different from data labeling or model evaluation?

Human beings want to help out other human beings, spread knowledge and might want to get recognition for it. Manually correcting (3 different) automation efforts seems like incredible monotone, unrewarding labour for a race to the bottom. Nobody should spend their time correcting AI models without compensation.